betaFeedback

WorksMathematics

Stability conditions on Lazarsfeld-Mukai bundles over abelian surfaces

Abstract

We study the stability of Lazarsfeld-Mukai bundles associated with smooth curves over abelian surfaces. When the general curve in ∣L∣|L| have Clifford dimension one and maximal gonality, for ρ≫0ρ\gg 0 we show that every component of the relative Brill-Noether loci Wd1(∣L∣)W^1_d(|L|) dominating the linear system parametrizes generically μLμ_L-stable rank-two Lazarsfeld-Mukai bundles. We also apply Bayer-Li wall-crossing to higher-rank bundles: if NS⁡(S)=Z⋅L\operatorname{NS}(S)=\mathbb Z\cdot L, AA is a globally generated complete linear series of degree dd on a smooth curve C∈∣L∣C\in|L|, h0(C,A)=r+1h^0(C,A)=r+1 with r>2r>2, and d≠g−1d\ne g-1, then the associated bundle EC,AE_{C,A} is μLμ_L-stable and LL-Gieseker-stable. The corresponding object EC,A∨[1]E_{C,A}^{\vee}[1] for d<g−1d<g-1, or EC,AE_{C,A} for d>g−1d>g-1, is stable in the adjacent chamber, as its reduction. These results connect dimension estimates for Brill-Noether loci with stability and wall-crossing on abelian surfaces.

Introduction

Let SS be a smooth projective abelian surface, let LL be ample, and write ∣L∣s|L|_s for its smooth connected curves, each of genus g=1+L2/2g=1+L^2/2. For such a curve CC, set Wd1(C)={A∈Pic⁡d(C):h0(C,A)≥2}W^1_d(C)=\{A\in\operatorname{Pic}^d(C):h^0(C,A)\ge2\}; the relative locus Wd1(∣L∣s)W^1_d(|L|_s) parametrizes pairs (C,A)(C,A) with A∈Wd1(C)A\in W^1_d(C). Its expected fiber dimension, when nonnegative, is the Brill-Noether number ρ(g,1,d)=g−2(g−d+1)=2d−g−2\rho(g,1,d)=g-2(g-d+1)=2d-g-2, so the expected relative dimension is g−2+ρ(g,1,d)g-2+\rho(g,1,d); a negative Brill-Noether number predicts an empty general fiber. The gonality is the least degree of a pencil, and the rank-two discussion assumes general curves have Clifford dimension one (their Clifford index is computed by a pencil) and maximal gonality k=⌊(g+3)/2⌋k=\lfloor(g+3)/2\rfloor. Knutsen, Lelli-Chiesa, and Mongardi study Severi varieties and Brill-Noether loci on general polarized abelian surfaces [6], while Bayer-Li describe Brill-Noether theory for curves in primitive systems on generic abelian surfaces [2]. Moretti’s recent work computes gonality and the dimensions of Wd1(C)W^1_d(C) for general curves in non-primitive systems on simple abelian surfaces, and derives Green’s conjecture in that setting [9].

The Lazarsfeld-Mukai construction associates a vector bundle on SS to a series on CC by extending its evaluation data from the curve to the surface; a complete base-point-free gdrg^r_d gives a bundle of rank r+1r+1. Its stability connects the geometry of this series with moduli spaces of sheaves: the slope stability compares μL(E)=c1(E)⋅L/rk⁡(E)\mu_L(E)=c_1(E)\cdot L/\operatorname{rk}(E) with subsheaf slopes, while Gieseker stability is defined using reduced Hilbert polynomials. Lazarsfeld’s foundational work on Brill-Noether-Petri theory for curves on K3K3 surfaces established a vector-bundle approach that informs later Lazarsfeld-Mukai methods [7]. Narayanan studies semistability of Lazarsfeld-Mukai bundles on abelian surfaces and stability for dominating components on associated Kummer surfaces [10].

For a complete base-point-free pencil on a smooth curve in an ample linear system, the associated LM bundle has rank two. We study possible destabilizing filtrations and use intersection-theoretic and parameter-space estimates to assess how non-stable bundles may contribute to the relative Brill-Noether locus. The Clifford-dimension and gonality hypotheses enter these estimates; a bound on the non-stable sublocus alone does not establish the expected dimension or reducedness of the full Brill-Noether locus.

The final part applies Bayer-Li wall-crossing to complete, globally generated series of higher dimension [2]. If SS has NS⁡(S)=Z⋅L\operatorname{NS}(S) = \mathbb{Z} \cdot L, C∈∣L∣C \in\lvert L\rvert is smooth, and AA is a globally generated line bundle of degree dd with h0(C,A)=r+1h^{0}(C,A) = r+1, for r>2r > 2, and d≠g−1d \ne g-1, then EC,AE_{C,A} is μL\mu_L-stable and LL-Gieseker-stable. Wall-crossing varies the stability condition across a wall in the derived category, where stability can change; this higher-rank argument is independent of the rank-two estimates. Bayer explains how Bridgeland wall-crossing recovers Lazarsfeld’s Brill-Noether theorem for curves on K3K3 surfaces and surveys applications of stability conditions on surfaces [1]. Feyzbakhsh’s effective restriction theorem gives a complementary criterion for stability of restrictions of sheaves to divisors [3].

§ 2 recalls the LM construction and its basic invariants; § 3 studies rank-two stability and parameter estimates; § 4 introduces the Bridgeland wall, states the Bayer-Li results used here, and proves the higher-rank stability theorem.

Lazarsfeld-Mukai bundles for abelian surfaces

Let AA be a complex abelian surface. For any coherent sheaf QQ on AA, the Euler characteristic is defined as the alternating sum of the cohomology dimensions

χ(A,Q):=∑i=02(−1)idim⁡Hi(A,Q).\chi(A,Q) := \sum_{i=0}^{2}(-1)^i \dim H^i(A,Q).

For AA an abelian surface, the tangent bundle has vanishing Chern classes c1(TA)=0c_1(T_A) = 0 and c2(TA)=0c_2(T_A) = 0, so the Todd class td⁡(A)\operatorname{td}(A) is trivial. Following [4], the Hirzebruch-Riemann-Roch theorem then yields

χ(A,Q)=∫Ach⁡(Q),\chi(A,Q) = \int_A \operatorname{ch}(Q),

where ch⁡(Q)\operatorname{ch}(Q) is the Chern character of QQ. In terms of the Chern classes r=rk⁡(Q)r = \operatorname{rk}(Q), c1=c1(Q)c_1 = c_1(Q), and c2=c2(Q)c_2 = c_2(Q), this becomes

χ(A,Q)=12c12−c2,\chi(A,Q) = \frac{1}{2}c_1^2-c_2,

with all integrals evaluated against the fundamental class of AA. In the special case where Q=LQ=L is a line bundle, the rank rr is 1 and c2(L)=0c_2(L)=0, so the Euler characteristic simplifies to

χ(A,L)=12c1(L)2.\chi(A,L)=\frac{1}{2}c_1(L)^2.

As in [5], we have that the Mukai vector of a coherent sheaf QQ on a surface AA is defined by applying the Chern character to the modified Todd class, namely

v(Q):=ch⁡(Q)td⁡(A).v(Q):=\operatorname{ch}(Q)\sqrt{\operatorname{td}(A)}.

On an abelian surface AA, the Todd class is trivial, so v(Q)=ch⁡(Q)v(Q)=\operatorname{ch}(Q) lies in H0(A,Z)⊕H2(A,Z)⊕H4(A,Z)H^0(A,\mathbb{Z})\oplus H^2(A,\mathbb{Z})\oplus H^4(A,\mathbb{Z}).

In terms of the Chern classes r=rk⁡(Q)r=\operatorname{rk}(Q), c1=c1(Q)c_1=c_1(Q), and c2=c2(Q)c_2=c_2(Q), the Mukai vector reads

v(Q)=(r,c1(Q),12c1(Q)2−c2(Q)).v(Q)=\left(r,c_1(Q),\frac{1}{2}c_1(Q)^2-c_2(Q)\right).

Derived equivalences and Fourier-Mukai transforms transport this vector by the induced cohomological isometry, and it is a fundamental invariant in the study of moduli spaces of sheaves on abelian surfaces. If Q=LQ=L is a line bundle on AA, then r=1r=1 and c2(L)=0c_2(L)=0, so the Mukai vector becomes

v(L)=(1,c1(L),12c1(L)2).v(L)=\left(1,c_1(L),\frac{1}{2}c_1(L)^2\right).

Let AA be a complex abelian surface and let QQ and RR be coherent sheaves on AA, with Mukai vectors

v(Q)=(v0(Q),v1(Q),v2(Q)),v(R)=(v0(R),v1(R),v2(R)),v(Q)=\left(v_0(Q),v_1(Q),v_2(Q)\right),\qquad v(R)=\left(v_0(R),v_1(R),v_2(R)\right),

lying in H0(A,Z)⊕H2(A,Z)⊕H4(A,Z)H^0(A,\mathbb{Z})\oplus H^2(A,\mathbb{Z})\oplus H^4(A,\mathbb{Z}). The Mukai pairing of v(Q)v(Q) and v(R)v(R) is the symmetric bilinear form defined as

⟨v(Q),v(R)⟩:=−∑i=02(−1)idim⁡Ext⁡Ai(Q,R).\langle v(Q),v(R)\rangle:=-\sum_{i=0}^{2}(-1)^i\dim\operatorname{Ext}^i_A(Q,R).

If one writes v(Q)=(r,c1,v2)v(Q)=(r,c_1,v_2), so v0=rv_0=r, v1=c1v_1=c_1, v2=12c12−c2v_2=\frac{1}{2}c_1^2-c_2, and similarly v(R)=(r′,c1′,v2′)v(R)=(r',c_1',v_2'), then the Mukai pairing reads explicitly

⟨v(Q),v(R)⟩=v1(Q)⋅v1(R)−v0(Q)v2(R)−v2(Q)v0(R),\langle v(Q),v(R)\rangle=v_1(Q)\cdot v_1(R)-v_0(Q)v_2(R)-v_2(Q)v_0(R),

where v1(Q)⋅v1(R)v_1(Q)\cdot v_1(R) denotes the intersection product ∫Av1(Q)∧v1(R)\int_A v_1(Q)\wedge v_1(R).

Let AA be a complex abelian surface and let MM be a coherent sheaf on AA with fixed Mukai vector v(M)=(r,c1,v2)v(M)=(r,c_1,v_2). Fix an ample divisor class HH on $A and consider the moduli space MHss(v(M))\mathcal{M}^{\mathrm{ss}}_H(v(M)) of HH-semistable sheaves on AA with Mukai vector v(M)v(M), in the sense of geometric invariant theory as developed in [5]. The open subscheme MHs(v(M))⊂MHss(v(M))\mathcal{M}^{\mathrm{s}}_H(v(M))\subset\mathcal{M}^{\mathrm{ss}}_H(v(M)) parametrizes HH-stable sheaves and, when this moduli space is nonempty, v(M)v(M) is primitive, and HH is generic for v(M)v(M), stable and semistable objects coincide and the moduli space is smooth, irreducible, and projective over C\mathbb{C}.

The expected dimension of the moduli space is governed by the deformation theory of the sheaf. For a stable sheaf EE with rk⁡(E)=r\operatorname{rk}(E)=r and Mukai vector v(E)v(E), the tangent space at [E][E] is identified with Ext⁡1(E,E)\operatorname{Ext}^{1}(E,E) and the obstruction space with Ext⁡2(E,E)\operatorname{Ext}^{2}(E,E). On an abelian surface, the trace-free obstruction space vanishes, whereas Ext⁡2(E,E)≃C\operatorname{Ext}^{2}(E,E)\simeq\mathbb{C}; the stable coarse moduli space therefore has dimension ⟨v(E),v(E)⟩+2\langle v(E),v(E)\rangle+2; see [5].

When MM is a line bundle, v(M)=(1,c1(M),12c1(M)2)v(M)=(1,c_{1}(M),\frac{1}{2}c_{1}(M)^{2}) and the Mukai self-pairing vanishes, hence the stable moduli space has dimension 2.

Let SS be a smooth complex projective abelian surface, let i:C↪Si:C\hookrightarrow S be a smooth connected curve, and denote as L=OS(C)L=\mathcal{O}_{S}(C) and g=1+L2/2g=1+L^{2}/2. Let AA be a globally generated line bundle of degree dd on CC, with h0(C,A)=n=r+1h^{0}(C,A)=n=r+1. Thus we use a complete gdrg^{r}_{d}, not an arbitrary point of Wdr(C)W^{r}_{d}(C) with possibly more sections. Evaluation on SS is surjective

0⟶FC,A⟶H0(C,A)⊗OS→ev⁡A,Si∗A⟶0,(1)0\longrightarrow F_{C,A}\longrightarrow H^{0}(C,A)\otimes\mathcal{O}_{S}\xrightarrow{\operatorname{ev}_{A,S}}i_{*}A\longrightarrow0, \tag*{(1)}

and the kernel is locally free. The Lazarsfeld-Mukai bundle associated to (C,A)(C,A) is EC,A=FC,A∨E_{C,A}=F_{C,A}^{\vee}. Adjunction gives ωC≃L∣C\omega_{C}\simeq L|_{C}, and dualizing (1) yields

0⟶H0(C,A)∨⊗OS⟶EC,A⟶i∗(ωC⊗A−1)⟶0.(2)0\longrightarrow H^{0}(C,A)^{\vee}\otimes\mathcal{O}_{S}\longrightarrow E_{C,A}\longrightarrow i_{*}(\omega_{C}\otimes A^{-1})\longrightarrow0. \tag*{(2)}

Both quotients are pushforwards from CC, not vector bundles on SS. In particular, EC,AE_{C,A} has a distinguished nn-dimensional subspace of sections; this need not be its entire space of sections.

Theorem 2.1 (Basic LM invariants). With the above assumptions, write E=EC,AE=E_{C,A} and B=ωC⊗A−1B=\omega_{C}\otimes A^{-1}. Then:

  • rk⁡(E)=n\operatorname{rk}(E)=n, det⁡(E)=L\det(E)=L, c2(E)=dc_{2}(E)=d, and χ(S,E)=g−1−d\chi(S,E)=g-1-d;

  • h2(S,E)=0h^{2}(S,E)=0. For the map γ:H0(C,A)⊗H1(S,OS)→H1(C,A)\gamma:H^{0}(C,A)\otimes H^{1}(S,\mathcal{O}_{S})\to H^{1}(C,A) induced by evaluation,

τ:=h1(S,E)=dim⁡ker⁡γ,0≤τ≤2n;\tau:=h^{1}(S,E)=\dim\ker\gamma,\qquad0\leq\tau\leq2n;
  • h0(S,E)=g−1−d+τ=n+h1(C,A)−rank⁡γ≥nh^{0}(S,E)=g-1-d+\tau=n+h^{1}(C,A)-\operatorname{rank}\gamma\geq n;

  • the distinguished sections generate EE away from CC. If the image of H0(S,E)→H0(C,B)H^{0}(S,E)\to H^{0}(C,B) generates BB away from a finite subset of CC, then EE is generated outside a finite subset of SS;

  • v(E)=(n,L,g−1−d)v(E)=(n,L,g-1-d) and v(E)2=L2−2n(g−1−d)v(E)^{2}=L^{2}-2n(g-1-d).

Proof. The character of i∗Ai_{*}A is (0,L,d+1−g)(0,L,d+1-g). Additivity in (1) and duality give the rank, determinant, Chern classes and Euler characteristic. The induced map H0(C,A)⊗H0(S,OS)→H0(C,A)H^{0}(C,A)\otimes H^{0}(S,\mathcal{O}_{S})\to H^{0}(C,A) is the identity. Hence H0(S,FC,A)=0H^{0}(S,F_{C,A})=0 and H1(S,FC,A)≃ker⁡γH^{1}(S,F_{C,A})\simeq\ker\gamma. Serre duality gives H2(S,E)=0H^{2}(S,E)=0 and h1(S,E)=h1(S,FC,A)h^{1}(S,E)=h^{1}(S,F_{C,A}). Now h1(S,OS)=2h^{1}(S,\mathcal{O}_{S})=2 and Riemann-Roch prove the cohomology formulas. The generation statements follow from (2); the Mukai square uses the convention fixed above. □\square

Notice that from the non-vanishing cohomology H1(S,OS)H^{1}(S,\mathcal{O}_{S}) the K3 formulas cannot be transferred unchanged to an abelian surface. In particular, neither h0(S,E)=nh^{0}(S,E)=n nor automatic generation along CC is asserted. For the restriction and Petri calculations, define the kernel on CC by

0→MA→H0(C,A)⊗OC→A→0.0 \to M_{A} \to H^{0}(C,A) \otimes\mathcal{O}_{C} \to A \to0.

Restricting (2.1) to CC and accounting for its Tor term gives

0→A⊗ωC−1→FC,A∣C→MA→0.(3)0 \to A \otimes\omega_{C}^{-1} \to F_{C,A}|_{C} \to M_{A} \to0. \tag*{(3)}

Equivalently, tensoring by BB gives

0→OC→FC,A∣C⊗B→MA⊗B→0.0 \to\mathcal{O}_{C} \to F_{C,A}|_{C} \otimes B \to M_{A} \otimes B \to0.

The Petri map μA:H0(C,A)⊗H0(C,B)→H0(C,ωC)\mu_{A}: H^{0}(C,A) \otimes H^{0}(C,B) \to H^{0}(C,\omega_{C}) has kernel H0(C,MA⊗B)H^{0}(C,M_{A} \otimes B). These are statements on the curve; they do not supply an isomorphism involving H0(S,EC,A⊗FC,A∨)H^{0}(S,E_{C,A} \otimes F_{C,A}^{\vee}).

Proposition 2.2 ([8], Prop. 3.1). Let EE and FF be torsion-free sheaves on SS. Then:

(1) If EE and FF are μH\mu_{H}-semistable with μH(E)>μH(F)\mu_{H}(E) > \mu_{H}(F), then Hom⁡(E,F)=0\operatorname{Hom}(E,F)=0.

(2) If EE and FF are μH\mu_{H}-stable with μH(E)=μH(F)\mu_{H}(E)=\mu_{H}(F) and 0≠ϕ∈Hom⁡(E,F)0 \ne\phi\in\operatorname{Hom}(E,F), then rk⁡(E)=rk⁡(F)\operatorname{rk}(E)=\operatorname{rk}(F) and ϕ\phi is injective and an isomorphism away from a finite subset of SS.

Remark 2.3. On an abelian surface SS since KS≅OSK_{S} \cong\mathcal{O}_{S} by Serre duality H2(S,E)≅Hom⁡(E,OS)∨H^{2}(S,E) \cong\operatorname{Hom}(E,\mathcal{O}_{S})^{\vee} and the first part of the above Proposition implies that, if EE is μH\mu_{H}-semistable and μH(E)>0\mu_{H}(E)>0 then h2(S,E)=0h^{2}(S,E)=0. Also if EE is μH\mu_{H}-semistable and μH(E)<0\mu_{H}(E)<0 then h0(S,E)=0h^{0}(S,E)=0.

Lemma 2.4 ([8], Lemma 3.2). Let E,Q∈Coh⁡(S)E,Q \in\operatorname{Coh}(S) be torsion-free sheaves with rk⁡(E)≥2\operatorname{rk}(E) \ge2 and rk⁡(Q)>0\operatorname{rk}(Q)>0. Suppose EE is globally generated outside a finite set of points, h2(S,E)=0h^{2}(S,E)=0 and there exists a surjective morphism ϕ:E→Q\phi:E \to Q. Then h0(S,Q∨∨)≥2h^{0}(S,Q^{\vee\vee}) \ge2.

Proof. As a quotient of EE, the sheaf QQ is globally generated outside a finite set. If rk⁡(Q)≥2\operatorname{rk}(Q) \ge2, then

h0(S,Q∨∨)≥h0(S,Q)≥2,h^{0}(S,Q^{\vee\vee}) \ge h^{0}(S,Q) \ge2,

as required.

Now suppose rk⁡(Q)=1\operatorname{rk}(Q)=1. Write Q=N⊗IξQ=N \otimes I_{\xi} where N=Q∨∨N=Q^{\vee\vee} is a line bundle and ξ\xi is zero-dimensional. Generation outside a finite subset implies that NN has a nonzero section and no fixed divisorial component. If h0(S,N)=1h^{0}(S,N)=1, its unique section has no divisorial zero locus and therefore trivializes NN. The nonzero composite E↠Q↪N≅OSE \twoheadrightarrow Q \hookrightarrow N \cong\mathcal{O}_{S} would contradict h2(S,E)=dim⁡Hom⁡(E,OS)=0h^{2}(S,E)=\dim\operatorname{Hom}(E,\mathcal{O}_{S})=0. Hence h0(S,N)≥2h^{0}(S,N) \ge2. No assertion of base-point-freeness is needed; a line bundle with no fixed divisorial component may have isolated base points. □

Stability of Lazarsfeld-Mukai bundles of rank 2

Let SS be a smooth, projective abelian surface and consider a line bundle L∈Amp(S)L \in Amp(S) such that a general curve C∈∣L∣sC \in\lvert L\rvert_s has genus gg, Clifford dimension 1 and maximal gonality k=⌊g+12⌋k = \left\lfloor\frac{g+1}{2} \right\rfloor. In this section we prove that, if CC is general in its linear system and ρ(g,1,d)>0\rho(g,1,d) > 0, the LM bundle associated with a general complete, base-point-free gd1g_d^1 on CC is μL\mu_L-stable.

Fix a rank-2 LM bundle E=EC,AE=E_{C,A} corresponding to a complete, base-point-free pencil A∈Wd1(C)A \in W_d^1(C) with C∈∣L∣sC \in\lvert L\rvert_s; we assume EE is not μL\mu_L-stable. In the case where EE is μL\mu_L-unstable (resp. properly μL\mu_L-semistable) we consider its HN filtration (resp. JH filtration) 0⊂M⊂E0 \subset M \subset E, which gives a short exact sequence

0→M→E→N⊗Iξ→0,(4)0 \to M \to E \to N \otimes I_{\xi} \to0, \tag*{(4)}

where MM and NN are two line bundles such that

μL(M)>μL(E)=g−1>μL(N)(resp. μL(M)=μL(E)=μL(N))\mu_L(M) > \mu_L(E) = g-1 > \mu_L(N) \quad(\text{resp. } \mu_L(M)=\mu_L(E)=\mu_L(N))

and IξI_{\xi} is the ideal sheaf of a 0-dimensional subscheme ξ⊂S\xi\subset S of length l=d−c1(N)⋅c1(M)l=d-c_1(N)\cdot c_1(M).

Lemma 3.1. If general curves in ∣L∣s\lvert L\rvert_s have Clifford dimension 1 and constant gonality kk, one has c1(M)⋅c1(N)≥k−4c_1(M)\cdot c_1(N) \ge k-4

Proof. Since g−1<μL(M)=c1(M)2+c1(N)⋅c1(M)g-1 < \mu_L(M)=c_1(M)^2+c_1(N)\cdot c_1(M) then if c1(M)2≤0c_1(M)^2 \le0 we get

g≤c1(M)2+c1(N)⋅c1(M)≤c1(N)⋅c1(M).g \le c_1(M)^2+c_1(N)\cdot c_1(M) \le c_1(N)\cdot c_1(M).

On the other hand, from hypothesis in the gonality

k−1≤g+12≤g.k-1 \le\frac{g+1}{2} \le g.

By putting together k−1≤c1(N)⋅c1(M)k-1 \le c_1(N)\cdot c_1(M) and the lemma holds. From now on we assume

c1(M)2>0.c_1(M)^2 > 0.

Since μL(M)>0\mu_L(M)>0 from Remark 2.3 it follows h2(S,M)=0h^2(S,M)=0 and χ(M)=h0(S,M)−h1(S,M)\chi(M)=h^0(S,M)-h^1(S,M).

Case h0(S,M)=0h^0(S,M)=0: From Riemann-Roch theorem χ(M)=−h1(S,M)=c1(M)2/2\chi(M)=-h^1(S,M)=c_1(M)^2/2 which gives a contradiction on the assumption.

Case h0(S,M)=1h^0(S,M)=1: In this case χ(M)=1−h1(S,M)=c1(M)2/2\chi(M)=1-h^1(S,M)=c_1(M)^2/2 from our assumption h1(S,M)=0h^1(S,M)=0 and c1(M)2=2c_1(M)^2=2 then

g−1<μL(M)=c1(M)2+c1(N)⋅c1(M)=2+c1(N)⋅c1(M),g-1 < \mu_L(M)=c_1(M)^2+c_1(N)\cdot c_1(M)=2+c_1(N)\cdot c_1(M),

and g−3<c1(N)⋅c1(M)g-3<c_1(N)\cdot c_1(M). From the hypothesis in the gonality

k−4≤g−3,k-4 \le g-3,

it follows that k−4<c1(N)⋅c1(M)k-4<c_1(N)\cdot c_1(M).

::: Case h0(S,M)≥2h^0(S,M) \ge2: From Lemma 2.4, we know that h0(S,N)≥2h^0(S,N) \ge2 then c1(N)c_1(N) is effective and by Nakai-Moishezon criterion since LL is ample 0<c1(L)⋅c1(N)=μL(N)0 < c_1(L) \cdot c_1(N) = \mu_L(N) then Remark 2.3 implies h2(S,N)=0h^2(S,N) = 0. Since M=L⊗N∨M = L \otimes N^\vee then 0=h2(S,N)=h0(S,N∨)=h0(S,M(−C))0 = h^2(S,N) = h^0(S,N^\vee) = h^0(S,M(-C)). By adjunction formula over CC we get

M∣C=(L⊗N∨)∣C≅ωC⊗N∨∣C,M|_C = (L \otimes N^\vee)|_C \cong\omega_C \otimes N^\vee|_C,

and from the cohomology of

0→M(−C)→M→M∣C→0,0 \to M(-C) \to M \to M|_C \to0,

it follows that 2≤h0(S,M)≤h0(C,M∣C)=h0(C,ωC⊗N∨∣C)=h1(C,N∣C)2 \le h^0(S,M) \le h^0(C,M|_C) = h^0(C,\omega_C \otimes N^\vee|_C) = h^1(C,N|_C). Similarly the exact sequence

0→M∨→N→N∣C→0,0 \to M^\vee\to N \to N|_C \to0,

gives 2≤h0(S,N)≤h0(C,N∣C)2 \le h^0(S,N) \le h^0(C,N|_C), then N∣CN|_C contributes to Cliff⁡(C)\operatorname{Cliff}(C),

Cliff⁡(N∣C)=c1(N)⋅(c1(N)+c1(M))−2h0(N∣C)+2=c1(N)2+c1(N)⋅c1(M)−2h0(N∣C)+2\begin{aligned} \operatorname{Cliff}(N|_C) = c_1(N) \cdot(c_1(N) + c_1(M)) - 2h^0(N|_C) + 2 \\ &= c_1(N)^2 + c_1(N) \cdot c_1(M) - 2h^0(N|_C) + 2 \end{aligned}

since h2(S,N)=0h^2(S,N) = 0 then h0(C,N∣C)≥h0(S,N)=χ(N)+h1(S,N)h^0(C,N|_C) \ge h^0(S,N) = \chi(N) + h^1(S,N) and

Cliff⁡(N∣C)≤c1(N)2+c1(N)⋅c1(M)−2χ(N)−2h1(S,N)+2=c1(N)2+c1(N)⋅c1(M)−2(c1(N)2/2)−2h1(S,N)+2=c1(N)⋅c1(M)−2h1(S,N)+2\begin{aligned} \operatorname{Cliff}(N|_C) \le c_1(N)^2 + c_1(N) \cdot c_1(M) - 2\chi(N) - 2h^1(S,N) + 2 \\ &= c_1(N)^2 + c_1(N) \cdot c_1(M) - 2(c_1(N)^2/2) - 2h^1(S,N) + 2 \\ &= c_1(N) \cdot c_1(M) - 2h^1(S,N) + 2 \end{aligned}

From Cliff⁡(N∣C)≥k−2\operatorname{Cliff}(N|_C) \ge k - 2 follows that

c1(N)⋅c1(M)≥k−4+2h1(S,N)≥k−4.c_1(N) \cdot c_1(M) \ge k - 4 + 2h^1(S,N) \ge k - 4.

As in [8], we use the following notation: Fix a nonnegative integer ll and a non-trivial, globally generated line bundle NN on SS such that, having defined M=L⊗N∨M = L \otimes N^\vee either μL(M)=μL(N)=g−1\mu_L(M) = \mu_L(N) = g - 1 or μL(M)>g−1>μL(N)\mu_L(M) > g - 1 > \mu_L(N). We consider the moduli stack EN,l\mathcal{E}_{N,l} parametrizing filtrations 0⊂M⊂E0 \subset M \subset E with [M]∈M(v(M))(C)[M] \in\mathcal{M}(v(M))(C) and [E/M]∈M(v(N⊗Iξ))(C)[E/M] \in\mathcal{M}(v(N \otimes I_\xi))(C), where l(ξ)=ll(\xi) = l.

Remark 3.2. The dimension dim⁡EN,l\dim\mathcal{E}_{N,l} can be computed by considering the natural morphism

p:EN,l→M(v(M))×M(v(N⊗Iξ))p : \mathcal{E}_{N,l} \to\mathcal{M}(v(M)) \times\mathcal{M}(v(N \otimes I_\xi))

sending a short exact sequence as in (4) to (M,N⊗Iξ)(M,N \otimes I_\xi). The fiber of pp at a CC-point (M,N⊗Iξ)(M,N \otimes I_\xi) is the quotient stack [Ext⁡1(N⊗Iξ,M)/Hom⁡(N⊗Iξ,M)][\operatorname{Ext}^1(N \otimes I_\xi,M)/\operatorname{Hom}(N \otimes I_\xi,M)] with trivial action. The dimension of the fiber is dim⁡Ext⁡1(N⊗Iξ,M)−dim⁡Hom⁡(N⊗Iξ,M)\dim\operatorname{Ext}^1(N \otimes I_\xi,M)-\dim\operatorname{Hom}(N \otimes I_\xi,M). From the duality on abelian varieties Ext⁡i(F,G)≅Ext⁡2−i(G,F)∨\operatorname{Ext}^i(F,G) \cong\operatorname{Ext}^{2-i}(G,F)^\vee then

dim⁡Ext⁡1(N⊗Iξ,M)=dim⁡Ext⁡0(N⊗Iξ,M)+dim⁡Ext⁡2(N⊗Iξ,M)−χ(N⊗Iξ,M)=dim⁡Ext⁡0(N⊗Iξ,M)+dim⁡Ext⁡0(M,N⊗Iξ)−χ(N⊗Iξ,M)\begin{aligned} \dim\operatorname{Ext}^1(N \otimes I_\xi,M) = \dim\operatorname{Ext}^0(N \otimes I_\xi,M) + \dim\operatorname{Ext}^2(N \otimes I_\xi,M) - \chi(N \otimes I_\xi,M) \\ &= \dim\operatorname{Ext}^0(N \otimes I_\xi,M) + \dim\operatorname{Ext}^0(M,N \otimes I_\xi) - \chi(N \otimes I_\xi,M) \end{aligned}

since Ext⁡0(F,G)≅Hom⁡(F,G)\operatorname{Ext}^{0}(F,G) \cong\operatorname{Hom}(F,G) and χ(F,G)=−⟨v(F),v(G)⟩\chi(F,G)=-\langle v(F),v(G)\rangle the dimension of the fiber is

dim⁡Hom⁡(M,N⊗Iξ)−χ(N⊗Iξ,M)=dim⁡Hom⁡(M,N⊗Iξ)+⟨v(N⊗Iξ),v(M)⟩.\dim\operatorname{Hom}(M,N\otimes I_{\xi})-\chi(N\otimes I_{\xi},M)=\dim\operatorname{Hom}(M,N\otimes I_{\xi})+\langle v(N\otimes I_{\xi}),v(M)\rangle.

we conclude that

dim⁡EN,l=2l+dim⁡Hom⁡(M,N⊗Iξ)+⟨v(N⊗Iξ),v(M)⟩.\dim\mathcal{E}_{N,l}=2l+\dim\operatorname{Hom}(M,N\otimes I_{\xi})+\langle v(N\otimes I_{\xi}),v(M)\rangle.

Remark 3.3. Let MM and NN be line bundles on the surface, and let ξ\xi be a finite subscheme of length ll, with associated ideal sheaf IξI_{\xi}. We denote by N⊗IξN\otimes I_{\xi} the sheaf obtained by tensoring NN with IξI_{\xi}. The goal is to compute the Mukai pairings, using the componentwise formula

⟨v,w⟩=v1⋅w1−v0w2−v2w0.\langle v,w\rangle=v_{1}\cdot w_{1}-v_{0}w_{2}-v_{2}w_{0}.

The Mukai vector of a line bundle MM is

v(M)=(1,c1(M),12c1(M)2).v(M)=\left(1,c_{1}(M),\frac{1}{2}c_{1}(M)^{2}\right).

Similarly, for NN one has

v(N)=(1,c1(N),12c1(N)2).v(N)=\left(1,c_{1}(N),\frac{1}{2}c_{1}(N)^{2}\right).

To compute the Mukai vector of N⊗IξN\otimes I_{\xi}, consider the short exact sequence

0→N⊗Iξ→N→N∣ξ→0.0\to N\otimes I_{\xi}\to N\to N|_{\xi}\to0.

The sheaf N∣ξN|_{\xi} has character (0,0,l)(0,0,l). Thus

ch⁡(N⊗Iξ)=ch⁡(N)−ch⁡(N∣ξ)=(1,c1(N),12c1(N)2)−(0,0,l)=(1,c1(N),12c1(N)2−l).\begin{aligned} \operatorname{ch}(N\otimes I_{\xi})&=\operatorname{ch}(N)-\operatorname{ch}(N|_{\xi})\\ &=\left(1,c_{1}(N),\frac{1}{2}c_{1}(N)^{2}\right)-(0,0,l)\\ &=\left(1,c_{1}(N),\frac{1}{2}c_{1}(N)^{2}-l\right). \end{aligned}

Putting v=v(M)v=v(M), the self-pairing is

⟨v(M),v(M)⟩=c1(M)2−c1(M)22−c1(M)22=0.\langle v(M),v(M)\rangle=c_{1}(M)^{2}-\frac{c_{1}(M)^{2}}{2}-\frac{c_{1}(M)^{2}}{2}=0.

Thus the Mukai self-pairing of a line bundle is

⟨v(M),v(M)⟩=0.\langle v(M),v(M)\rangle=0.

Let v=v(N⊗Iξ)v=v(N\otimes I_{\xi}). Its rank remains one, not 1−l1-l. Then

⟨v(N⊗Iξ),v(N⊗Iξ)⟩=c1(N)2−2(c1(N)22−l)=2l.\langle v(N\otimes I_{\xi}),v(N\otimes I_{\xi})\rangle=c_{1}(N)^{2}-2\left(\frac{c_{1}(N)^{2}}{2}-l\right)=2l.

Therefore the self-pairing of the twisted sheaf is

⟨v(N⊗Iξ),v(N⊗Iξ)⟩=2l.\langle v(N\otimes I_{\xi}),v(N\otimes I_{\xi})\rangle=2l.

Now take

v(M)=(1,c1,12c12),v(N⊗Iξ)=(1,c1′,12(c1′)2−l),v(M)=\left(1,c_{1},\frac{1}{2}c_{1}^{2}\right),\qquad v(N\otimes I_{\xi})=\left(1,c_{1}^{\prime},\frac{1}{2}(c_{1}^{\prime})^{2}-l\right),

with c1=c1(M)c_1=c_1(M), c1′=c1(N)c_1'=c_1(N). The Mukai pairing is

⟨v(M),v(N⊗Iξ)⟩=c1(M)⋅c1(N)−(c1(N)22−l)−c1(M)22=c1(M)⋅c1(N)−c1(M)2+c1(N)22+l.\begin{aligned} \langle v(M),v(N\otimes I_\xi)\rangle=c_1(M)\cdot c_1(N)-\left(\frac{c_1(N)^2}{2}-l\right)-\frac{c_1(M)^2}{2} \\ &= c_1(M)\cdot c_1(N)-\frac{c_1(M)^2+c_1(N)^2}{2}+l. \end{aligned}

In terms of Chern classes, this is

⟨v(M),v(N⊗Iξ)⟩=c1(M)⋅c1(N)−c1(M)2+c1(N)22+l.(5)\langle v(M),v(N\otimes I_\xi)\rangle=c_1(M)\cdot c_1(N)-\frac{c_1(M)^2+c_1(N)^2}{2}+l. \tag*{(5)}

In particular, when l=0l=0, this recovers ⟨v(M),v(N)⟩\langle v(M),v(N)\rangle. For these stable rank-one sheaves, the coarse moduli dimensions are

dim⁡MH(v(M))=2,dim⁡MH(v(N⊗Iξ))=2l+2.\dim M_H(v(M))=2,\qquad\dim M_H(v(N\otimes I_\xi))=2l+2.

Their moduli stacks have dimensions 11 and 2l+12l+1, respectively. These dimensions follow from v2+2v^2+2 for coarse spaces and v2+1v^2+1 for stable stacks.

Remark 3.4. Consider the natural morphism

q:EN,l⟶M(v(E))q:\mathcal{E}_{N,l}\longrightarrow\mathcal{M}(v(E))

mapping an exact sequence as 3.1 to [E][E]. Denote as P~N,l=Im⁡(q)\widetilde{\mathcal{P}}_{N,l}=\operatorname{Im}(q) the closure of the image of qq and consider the open substack PN,l\mathcal{P}_{N,l} inside P~N,l\widetilde{\mathcal{P}}_{N,l} whose points correspond to vector bundles EE satisfying h2(S,E)=0h^2(S,E)=0 and h1(S,E)=τh^1(S,E)=\tau, with τ\tau fixed. This last condition defines a locally closed cohomology stratum, not necessarily an open substack; we use PN,l\mathcal{P}_{N,l} for that stratum.

Consider the Grassmann bundle GN,l→PN,l\mathcal{G}_{N,l}\to\mathcal{P}_{N,l} with fiber over a point [E][E] equal to G(2,H0(S,E))G(2,H^0(S,E)). By Riemann-Roch and h2(S,E)=0h^2(S,E)=0, h0(S,E)=g−1−d+τh^0(S,E)=g-1-d+\tau, so dim⁡G(2,H0(S,E))=2(g−3−d+τ)\dim G(2,H^0(S,E))=2(g-3-d+\tau) where g−1−d+τ≥2g-1-d+\tau\geq2 and if PN,l\mathcal{P}_{N,l} is not empty then

dim⁡GN,l=dim⁡PN,l+2(g−3−d+τ).\dim\mathcal{G}_{N,l}=\dim\mathcal{P}_{N,l}+2(g-3-d+\tau).

In order to estimate τ\tau by the cohomology of the filtration. We use the long exact cohomology sequences of (3.1) and 0→N⊗Iξ→N→N∣ξ→00\to N\otimes I_\xi\to N\to N|_\xi\to0. Put Q=N⊗IξQ=N\otimes I_\xi. Since μL(M)≥g−1>0\mu_L(M)\geq g-1>0, Serre duality gives H2(S,M)=0H^2(S,M)=0. For the connecting map δ:H0(S,Q)→H1(S,M)\delta:H^0(S,Q)\to H^1(S,M), the first sequence gives

0→coker⁡δ→H1(S,E)→H1(S,Q)→0.0\to\operatorname{coker}\delta\to H^1(S,E)\to H^1(S,Q)\to0.

Define the rank of restriction to the length-ll subscheme by

rξ=rank⁡(H0(S,N)→H0(ξ,N∣ξ)).r_\xi=\operatorname{rank}\left(H^0(S,N)\to H^0(\xi,N|_\xi)\right).

The second sequence, together with h0(ξ,N∣ξ)=lh^0(\xi,N|_\xi)=l and the vanishing of higher cohomology of N∣ξN|_\xi, gives

h0(S,Q)=h0(S,N)−rξ,h1(S,Q)=h1(S,N)+l−rξ.h^0(S,Q)=h^0(S,N)-r_\xi,\qquad h^1(S,Q)=h^1(S,N)+l-r_\xi.

Consequently the chosen extension satisfies the exact identity

τ=h1(S,M)+h1(S,N)+l−rξ−rank⁡δ.(6)\tau=h^1(S,M)+h^1(S,N)+l-r_\xi-\operatorname{rank}\delta. \tag*{(6)}

Using 0≤rank⁡δ≤h1(S,M)0 \le\operatorname{rank}\delta\le h^{1}(S,M) yields

h1(S,N)+l−rξ≤τ≤h1(S,M)+h1(S,N)+l−rξ.h^{1}(S,N)+l-r_{\xi} \le\tau\le h^{1}(S,M)+h^{1}(S,N)+l-r_{\xi}.

For a complete LM pencil, τ≤4\tau\le4 follows from the original evaluation sequence: H1(S,E∨)H^{1}(S,E^{\vee}) embeds into H0(C,A)⊗H1(S,OS)H^{0}(C,A)\otimes H^{1}(S,\mathcal{O}_{S}), a four-dimensional space, and Serre duality identifies its dimension with h1(S,E)h^{1}(S,E). Also h2(S,E)=0h^{2}(S,E)=0, χ(S,E)=g−1−d\chi(S,E)=g-1-d, and h0(S,E)≥2h^{0}(S,E)\ge2 imply τ≥d+3−g\tau\ge d+3-g. Thus the estimate used below is

τ≤min⁡{4,h1(S,M)+h1(S,N)+l−rξ}.(7)\tau\le\min\{4,h^{1}(S,M)+h^{1}(S,N)+l-r_{\xi}\}. \tag*{(7)}

with lower bound τ≥max⁡{0,d+3−g,h1(S,N)+l−rξ}\tau\ge\max\{0,d+3-g,h^{1}(S,N)+l-r_{\xi}\}.

This bound uses only the filtration and completeness of the pencil; it does not assume that both endpoints are ample. Since NN is nontrivial and globally generated, h2(S,N)=0h^{2}(S,N)=0. Hence the endpoint terms can be computed by Riemann–Roch:

h1(S,M)=h0(S,M)−12c1(M)2,h1(S,N)=h0(S,N)−12c1(N)2.h^{1}(S,M)=h^{0}(S,M)-\frac{1}{2}c_{1}(M)^{2}, \qquad h^{1}(S,N)=h^{0}(S,N)-\frac{1}{2}c_{1}(N)^{2}.

The remaining defect l−rξl-r_{\xi} records the failure of ξ\xi to impose independent conditions on sections of NN. Global generation gives rξ≥1r_{\xi}\ge1 if l>0l>0, but does not by itself give rξ=lr_{\xi}=l. The rank of δ\delta depends on the extension class, so the identity (6) need not be constant when only MM, NN, and ll are fixed.

The term dim⁡Hom⁡(M,N⊗Iξ)\dim\operatorname{Hom}(M,N\otimes I_{\xi}) has a different role: by Serre duality it equals dim⁡Ext⁡2(N⊗Iξ,M)\dim\operatorname{Ext}^{2}(N\otimes I_{\xi},M), and hence contributes to the extension-stack dimension in Remark 3.2. It also bounds the tangent dimension of the Quot fiber when the filtration is forgotten. It is not the rank of δ\delta and cannot replace that rank in (3.3). On locally closed strata where these dimensions and τ\tau are constant and base change holds, the extension contribution and the Grassmann contribution 2(g−3−d+τ)2(g-3-d+\tau) can be combined. Substituting (7) then gives the refined upper bound in the parameter calculation below.

Remark 3.5. Consider the rational map

hN,l:GN,l→Wd1(∣L∣),h_{N,l}:G_{N,l}\to W^{1}_{d}(|L|),

mapping a general point (E,Λ)(E,\Lambda) to the pair (CΛ,AΛ)(C_{\Lambda},A_{\Lambda}), where CΛC_{\Lambda} is the degeneracy locus of the evaluation map ev⁡Λ:Λ⊗OS→E\operatorname{ev}_{\Lambda}:\Lambda\otimes\mathcal{O}_{S}\to E, which is injective for a general Λ∈G(2,H0(S,E))\Lambda\in G(2,H^{0}(S,E)) and the fiber of hN,lh_{N,l} at (C,A)(C,A) is the quotient stack

[P(Hom⁡(E(C,A),ωC⊗A∨))∘/Aut⁡(E(C,A))].[\mathbb{P}(\operatorname{Hom}(E_{(C,A)},\omega_{C}\otimes A^{\vee}))^{\circ}/\operatorname{Aut}(E_{(C,A)})].

As in [8] any fibre of hN,lh_{N,l} has dimension −1-1.

Proposition 3.6. Assume that PN,lP_{N,l} is non-empty and let WW be an irreducible component of WN,lW_{N,l}. Then

dim⁡(W)≤g+d+8−k,\dim(W)\le g+d+8-k,

where kk is the gonality of any curve in ∣L∣s|L|_{s}.

Proof. The slope ordering of MM and NN gives

dim⁡Hom⁡(M,N⊗Iξ)={1,M≅N, ξ=∅0,otherwise\dim\operatorname{Hom}(M,N\otimes I_{\xi}) = \begin{cases} 1, & M\cong N,\ \xi=\varnothing\\ 0, & \text{otherwise} \end{cases}

Indeed, strict inequality of slopes forces Hom⁡(M,N⊗Iξ)=0\operatorname{Hom}(M,N\otimes I_{\xi})=0. For equal slopes, a nonzero morphism M→NM\to N defines an effective divisor of zero intersection with the ample class LL, so M≅NM\cong N. In that case Hom⁡(M,N⊗Iξ)=H0(S,Iξ)\operatorname{Hom}(M,N\otimes I_{\xi})=H^0(S,I_{\xi}), which has dimension one for l=0l=0 and zero for l>0l>0. It follows from Remark 3.2 that the dimension of the fibers of pp is constant and equals ⟨v(N⊗Iξ),v(M)⟩\langle v(N\otimes I_{\xi}),v(M)\rangle, unless M≅NM\cong N and l=0l=0, in which case it is ⟨v(N⊗Iξ),v(M)⟩+1\langle v(N\otimes I_{\xi}),v(M)\rangle+1 then

dim⁡EN,l=2l+dim⁡Hom⁡(M,N⊗Iξ)+⟨v(N⊗Iξ),v(M)⟩.\dim\mathcal{E}_{N,l}=2l+\dim\operatorname{Hom}(M,N\otimes I_{\xi})+\langle v(N\otimes I_{\xi}),v(M)\rangle.

It is well known ([5 Proposition 2.2.8]) that the fiber of qq identifies with Quot⁡S(E,P)\operatorname{Quot}_{S}(E,P) and that given an element [φ:E→N⊗Iξ][\varphi:E\to N\otimes I_{\xi}] then

dim⁡[φ]Quot⁡S(E,P)≤dim⁡Hom⁡(K,N⊗Iξ)\dim_{[\varphi]}\operatorname{Quot}_{S}(E,P)\leq\dim\operatorname{Hom}(K,N\otimes I_{\xi})

where K=ker⁡φK=\ker\varphi and the fiber is zero-dimensional when dim⁡Hom⁡(M,N⊗Iξ)=0\dim\operatorname{Hom}(M,N\otimes I_{\xi})=0; the exceptional case M≅NM\cong N, l=0l=0 must be treated separately. From Remark 3.4 dim⁡PN,l≤dim⁡EN,l\dim\mathcal{P}_{N,l}\leq\dim\mathcal{E}_{N,l} and if PN,l≠∅\mathcal{P}_{N,l}\neq\varnothing

dim⁡GN,l≤2l+⟨v(N⊗Iξ),v(M)⟩+dim⁡Hom⁡(M,N⊗Iξ)+2(g−3−d+τ)=3l+2c1(M)⋅c1(N)−(g−1)+dim⁡Hom⁡(M,N⊗Iξ)+2(g−3−d+τ)=g+d−5−c1(M)⋅c1(N)+2τ+dim⁡Hom⁡(M,N⊗Iξ)≤g+d−5−c1(M)⋅c1(N)+2min⁡{4,h1(S,M)+h1(S,N)+l−rξ}+dim⁡Hom⁡(M,N⊗Iξ)≤g+d+8−k.\begin{aligned} \dim\mathcal{G}_{N,l} &\leq2l+\langle v(N\otimes I_{\xi}),v(M)\rangle\\ &\quad+\dim\operatorname{Hom}(M,N\otimes I_{\xi})+2(g-3-d+\tau) \\ &=3l+2c_1(M)\cdot c_1(N)-(g-1) \\ &\quad+\dim\operatorname{Hom}(M,N\otimes I_{\xi})+2(g-3-d+\tau) \\ &=g+d-5-c_1(M)\cdot c_1(N)+2\tau+\dim\operatorname{Hom}(M,N\otimes I_{\xi}) \\ &\leq g+d-5-c_1(M)\cdot c_1(N) \\ &\quad+2\min\{4,h^1(S,M)+h^1(S,N)+l-r_{\xi}\} \\ &\quad+\dim\operatorname{Hom}(M,N\otimes I_{\xi}) \\ &\leq g+d+8-k. \end{aligned}

The final inequality assumes τ≤4\tau\leq4, dim⁡Hom⁡(M,N⊗Iξ)≤1\dim\operatorname{Hom}(M,N\otimes I_{\xi})\leq1, and c1(M)⋅c1(N)≥k−4c_1(M)\cdot c_1(N)\geq k-4. where we have used that l=d−c1(M)⋅c1(N)l=d-c_1(M)\cdot c_1(N), c1(E)=c1(M)+c1(N)c_1(E)=c_1(M)+c_1(N) and c1(E)2=2g−2c_1(E)^2=2g-2. □

Theorem 3.7. Assume that general curves in ∣L∣s|L|_s have Clifford dimension 1 and maximal gonality k=⌊g+32⌋k=\left\lfloor\frac{g+3}{2}\right\rfloor. If ρ(g,1,d)>20\rho(g,1,d)>20, any dominating component of Wd1(∣L∣)W^1_d(|L|) corresponds to μL\mu_L-stable LM bundles. In particular, if C∈∣L∣sC\in|L|_s is general, the variety Wd1(∣L∣)W^1_d(|L|) is reduced and has the expected dimension ρ(g,1,d)\rho(g,1,d).

Proof. From the above Proposition the dimension of any non-stable component WW of Wd1(∣L∣)\mathcal{W}^{1}_{d}(|L|) is bounded by

dim⁡W≤g+d+8−k≤g+d+8−g+22=g2+d+7.\dim W \leq g+d+8-k \leq g+d+8-\frac{g+2}{2}=\frac{g}{2}+d+7.

On the other hand, any dominating component of Wd1(∣L∣)\mathcal{W}^{1}_{d}(|L|) must have dimension at least

dim⁡W≥dim⁡∣L∣s+ρ(g,1,d)=g−2+ρ(g,1,d)=g−2+(2d−g−2)=2d−4.\dim W \geq\dim|L|_{s}+\rho(g,1,d)=g-2+\rho(g,1,d)=g-2+(2d-g-2)=2d-4.

Then

2d−4≤dim⁡W≤d+g2+7,2d-4\leq\dim W\leq d+\frac{g}{2}+7,

and d≤g+22+10d\leq\frac{g+2}{2}+10. However when ρ>20\rho>20

d=ρ+g+22>g2+11.d=\frac{\rho+g+2}{2}>\frac{g}{2}+11.

Bridgeland Stability

Let SS be a smooth complex projective abelian surface with

NS⁡(S)=Z⋅L,L ample.(8)\operatorname{NS}(S)=\mathbb{Z}\cdot L,\qquad L\ \text{ample}. \tag*{(8)}

For a smooth connected curve i:C↪Si:C\hookrightarrow S in ∣L∣|L|, put g=1+L2/2g=1+L^{2}/2. Let AA be globally generated of degree dd, and write

U=H0(C,A),n=r+1=dim⁡U,r>2,χC=d+1−g.U=H^{0}(C,A),\qquad n=r+1=\dim U,\qquad r>2,\qquad\chi_{C}=d+1-g.

Thus rr denotes the dimension of the complete gdrg^{r}_{d}, not the bundle rank. Set E=EC,AE=E_{C,A}, F=E∨F=E^{\vee}, Q=i∗AQ=i_{*}A, and QB=i∗BQ_{B}=i_{*}B, where B=ωC⊗A−1B=\omega_{C}\otimes A^{-1}. The defining sequences are (2.1) and (2.2). By Theorem 2.1,

v(Q)=(0,L,χC),v(QB)=(0,L,−χC),v(E)=(n,L,−χC),v(F[1])=(−n,L,χC).(9)v(Q)=(0,L,\chi_{C}),\qquad v(Q_{B})=(0,L,-\chi_{C}),\qquad v(E)=(n,L,-\chi_{C}),\qquad v(F[1])=(-n,L,\chi_{C}). \tag*{(9)}

Assume χC≠0\chi_{C}\ne0 and choose the negative integer χ=−∣χC∣\chi=-|\chi_{C}|. Following [2 Definition 3.2], set

o=(1,0,0),R=⌊L2−2χ⌋,wk=(k−R,L,χ).(10)o=(1,0,0),\qquad R=\left\lfloor\frac{L^{2}}{-2\chi}\right\rfloor,\qquad w_{k}=(k-R,L,\chi). \tag*{(10)}

The plane Wχ=span⁡R{o,w0}W_{\chi}=\operatorname{span}_{\mathbb{R}}\{o,w_{0}\} determines the relevant wall. Let σ0\sigma_{0} be a geometric Bridgeland stability condition on it and let P0\mathcal{P}_{0} be the abelian category of σ0\sigma_{0}-semistable objects of the common phase of OS\mathcal{O}_{S} and wkw_{k}. Here geometric conditions are constructed on the tilted hearts Coh⁡βS\operatorname{Coh}^{\beta}S; their existence and the required wall-and-chamber properties are recalled in [2 Theorem 2.1 and Proposition 2.2], with the large-volume identifications in [2 Proposition 2.4].

Choose a nearby generic σ+\sigma_{+} with ν+(OS)<ν+(wk)\nu_{+}(\mathcal{O}_{S})<\nu_{+}(w_{k}). For a fixed finite range containing all classes used below, σ+\sigma_{+} can be chosen simultaneously in the ordinary Gieseker chamber for nonnegative rank and the dual Gieseker chamber for negative rank [2 Proposition 3.1].

In particular, stability of F[1]=E∨[1]F[1] = E^{\vee}[1] in the latter chamber is equivalent to Gieseker stability of EE. The description of stable factors and the adjacent-chamber criterion [2, Lemma 3.3, Proposition 3.4 and Corollary 3.5} give, for T∈P0T \in\mathcal{P}_0 of class wkw_k,

T is σ+-stable⟺Hom⁡(T,P)=0for all P∈Pic⁡0(S).T\text{ is }\sigma_+\text{-stable}\quad\Longleftrightarrow\quad\operatorname{Hom}(T,P)=0\quad\text{for all }P\in\operatorname{Pic}^0(S).

All exact sequences on the wall below are taken in P0\mathcal{P}_0, not necessarily in Coh⁡(S)\operatorname{Coh}(S).

For T∈Db(S)T \in D^b(S), write Hj(T)=Hom⁡(OS,T[j])H^j(T)=\operatorname{Hom}(\mathcal{O}_S,T[j]), hj(T)=dim⁡Hj(T)h^j(T)=\dim H^j(T), and set

Mkh={T∈Mσ+s(wk):h0(T)=h}.M_k^h=\{T\in M_{\sigma_+}^s(w_k):h^0(T)=h\}.

The number hh is a hypercohomology dimension and not necessarily equal to nn. For example, h0(F[1])=h1(S,E)h^0(F[1])=h^1(S,E) by Serre duality. Bayer-Li’s extension calculation gives dim⁡Ext⁡1(T,OS)=h−χ\dim\operatorname{Ext}^1(T,\mathcal{O}_S)=h-\chi for T∈MkhT\in M_k^h [2 Lemma 4.1].

Lemma 4.1 (Bayer-Li, Lemma 4.2). Let T∈MkhT\in M_k^h.

(a) For W⊆H0(T)W\subseteq H^0(T), evaluation gives an exact sequence in P0\mathcal{P}_0

0⟶OS⊗W⟶T⟶T′⟶0.0\longrightarrow\mathcal{O}_S\otimes W\longrightarrow T\longrightarrow T'\longrightarrow0.

Then T′T' is σ+\sigma_+-stable, has class wk−dim⁡Ww_{k-\dim W}, and h0(T′)≥dim⁡W+χh^0(T')\geq\dim W+\chi.

(b) For V⊆Ext⁡1(T,OS)V\subseteq\operatorname{Ext}^1(T,\mathcal{O}_S), the extension defined by the inclusion of VV gives an exact sequence

0⟶OS⊗V∨⟶T~⟶T⟶0.0\longrightarrow\mathcal{O}_S\otimes V^\vee\longrightarrow\widetilde{T}\longrightarrow T\longrightarrow0.

Then T~\widetilde{T} is σ+\sigma_+-stable, has class wk+dim⁡Vw_{k+\dim V}, and h0(T~)≥dim⁡Vh^0(\widetilde{T})\geq\dim V.

For the proof, see [2 Lemma 4.2].

The reduced object is defined by the unique exact sequence

0⟶J⟶T⟶Tred⟶0,Hom⁡(OS,Tred)=0,0\longrightarrow J\longrightarrow T\longrightarrow T_{\mathrm{red}}\longrightarrow0,\qquad\operatorname{Hom}(\mathcal{O}_S,T_{\mathrm{red}})=0,

where JJ is an iterated extension of copies of OS\mathcal{O}_S [2 Lemma 3.6 and Definition 3.7]. It is obtained by repeated evaluation quotients; one quotient need not remove all sections since h1(S,OS)=2h^1(S,\mathcal{O}_S)=2. The construction excludes OS\mathcal{O}_S as a subobject of TredT_{\mathrm{red}}, not as an arbitrary Jordan-Hölder factor.

Corollary 4.2 (Bayer-Li, Corollary 4.3). For T∈MkhT\in M_k^h, its reduction TredT_{\mathrm{red}} is σ+\sigma_+-stable and belongs to Mkred0M_{k_{\mathrm{red}}}^0 for some 0≤kred≤k0\leq k_{\mathrm{red}}\leq k.

For the proof, see [2 Corollary 4.3].

Theorem 4.3 (LM stability via Bayer-Li). Let SS be a smooth complex projective abelian surface with NS⁡(S)=Z⋅L\operatorname{NS}(S)=\mathbb{Z}\cdot L, where LL is ample. Let C∈∣L∣C\in|L| be smooth and connected, put g=1+L2/2g=1+L^2/2, and let AA be a globally generated line bundle of degree dd on CC with h0(C,A)=r+1h^0(C,A)=r+1 and r>2r>2. If d≠g−1d\ne g-1, then EC,AE_{C,A} is μL\mu_L-stable and LL-Gieseker-stable. Moreover, for the adjacent condition σ+\sigma_+ described above,

T={EC,A∨[1],d<g−1,EC,A,d>g−1.T= \begin{cases} E_{C,A}^{\vee}[1], & d<g-1,\\ E_{C,A}, & d>g-1. \end{cases}

is σ+\sigma_+-stable, and TredT_{\mathrm{red}} is σ+\sigma_+-stable with H0(Tred)=0H^0(T_{\mathrm{red}})=0.

Proof. Both QQ and QBQ_B are Gieseker-stable: a proper nonzero subsheaf of a line bundle on the integral curve CC has a zero-dimensional quotient of positive length and hence smaller reduced Hilbert polynomial. The chamber identifications above therefore apply to the sheaf with negative Euler characteristic.

If d<g−1d<g-1, then Q∈MRnQ\in M^n_R. Rotating the evaluation sequence (1) gives the exact sequence in P0\mathcal{P}_0

0⟶OS⊗U⟶Q⟶F[1]⟶0.0\longrightarrow\mathcal{O}_S\otimes U\longrightarrow Q\longrightarrow F[1]\longrightarrow0.

Lemma 4.1(a) implies that F[1]=E∨[1]F[1]=E^\vee[1] is σ+\sigma_+-stable of class wR−nw_{R-n}. The dual Gieseker chamber then gives Gieseker stability of EE.

If d>g−1d>g-1, then QBQ_B has class wRw_R and is σ+\sigma_+-stable. Serre duality and adjunction identify

Ext⁡1(QB,OS)≃H1(C,B)∨≃H0(C,A)=U.\operatorname{Ext}^1(Q_B,\mathcal{O}_S)\simeq H^1(C,B)^\vee\simeq H^0(C,A)=U.

Under this identification, the class of the dual evaluation sequence (2) corresponds to the identity on UU, as follows by dualizing (1). Thus Lemma 4.1(b), applied to the full space V=UV=U, gives σ+\sigma_+-stability of EE, of class wR+nw_{R+n}. The ordinary Gieseker chamber gives its Gieseker stability.

In either case EE is slope-semistable. If a proper saturated subsheaf of rank aa, 0<a<n0<a<n, had the same slope, its first Chern class would be mLmL for an integer mm, and equality of slopes would give mL2/a=L2/nmL^2/a=L^2/n, hence mn=amn=a, which is impossible. Therefore EE is μL\mu_L-stable. Finally, Corollary 4.2 applied to the stable object TT gives the assertion about its reduction. □

The preceding theorem establishes stability of the Lazarsfeld-Mukai bundle on SS. The following corollary combines that result with Feyzbakhsh’s effective restriction criterion, see [3]. We first recall the normalized discriminant and then state Proposition 4.6 in the notation of the present setting.

Let GG be a μL\mu_L-stable reflexive sheaf of rank q>1q>1 on SS. We use Feyzbakhsh’s normalized discriminant

Δ~(G):=(c1(G)⋅Lrk⁡(G)L2)2−2ch⁡2(G)rk⁡(G)L2.\widetilde{\Delta}(G):=\left(\frac{c_1(G)\cdot L}{\operatorname{rk}(G)L^2}\right)^2-\frac{2\operatorname{ch}_2(G)}{\operatorname{rk}(G)L^2}.

Theorem 4.4 ([3], Prop. 4.6). Let (S,L)(S,L) be a smooth polarized projective surface such that L2L^2 divides L⋅L′L\cdot L' for every L′∈Pic⁡(S)L'\in\operatorname{Pic}(S). Let GG be a μL\mu_L-stable reflexive sheaf of rank q>1q>1 satisfying

gcd⁡(L2rk⁡(G),c1(G)⋅L)=L2.\gcd\left(L^2\operatorname{rk}(G),c_1(G)\cdot L\right)=L^2.

Define

m(G)±=min⁡{m∈N:m′L2rk⁡(G)−mc1(G)⋅L=±L2 for some m′∈Z},m(G)_{\pm}=\min\left\{m\in\mathbb{N}:m'L^2\operatorname{rk}(G)-mc_1(G)\cdot L=\pm L^2\text{ for some }m'\in\mathbb{Z}\right\},
δ(G)=1qmin⁡{1m(G)+,1m(G)−}.\delta(G)=\frac{1}{q}\min\left\{\frac{1}{m(G)_+},\frac{1}{m(G)_-}\right\}.

For any irreducible divisor D∈∣kL∣D \in|kL|, the restriction G∣DG|_D is μ\mu-stable if

k>Δ~(G)δ(G)+δ(G),k > \frac{\widetilde{\Delta}(G)}{\delta(G)}+\delta(G),

and is μ\mu-semistable if the strict inequality is replaced by ≥\geq. In particular, G∣DG|_D is μ\mu-stable if

k>q(q−1)Δ~(G)+1q(q−1),k > q(q-1)\widetilde{\Delta}(G)+\frac{1}{q(q-1)},

and μ\mu-semistable if >> is replaced by ≥\geq.

In the application below, the Picard-rank-one hypothesis ensures the divisibility condition.

Corollary 4.5. Use the notation of Theorem 4.3, and assume d<g−1d<g-1. Thus U=H0(C,A)U=H^0(C,A), n=r+1n=r+1, E=EC,AE=E_{C,A}, and F=E∨F=E^\vee. If r>2r>2, then F∣CF|_C is μL∣C\mu_{L|_C}-stable.

Proof. The preceding theorem gives μL\mu_L-stability of EE, and hence of its dual FF, which is locally free. Recall that χC=d+1−g\chi_C=d+1-g and ch⁡(F)=(n,−L,−χC)\operatorname{ch}(F)=(n,-L,-\chi_C). Using L2=2(g−1)L^2=2(g-1), the normalized discriminant becomes

Δ~(F)=(c1(F)⋅LnL2)2−2ch⁡2(F)nL2=1r+1(dg−1−rr+1).\widetilde{\Delta}(F)=\left(\frac{c_1(F)\cdot L}{nL^2}\right)^2-\frac{2\operatorname{ch}_2(F)}{nL^2} =\frac{1}{r+1}\left(\frac{d}{g-1}-\frac{r}{r+1}\right).

Because NS⁡(S)=Z⋅L\operatorname{NS}(S)=\mathbb{Z}\cdot L, every divisor class is an integer multiple of LL, so the divisibility condition holds. For FF, the other hypothesis is satisfied since gcd⁡[L2rk⁡(F),c1(F)⋅L]=gcd⁡(nL2,−L2)=L2\gcd[L^2\operatorname{rk}(F),c_1(F)\cdot L]=\gcd(nL^2,-L^2)=L^2. Thus Proposition 4.6 applies to C∈∣L∣C\in|L| with k=1k=1, and its strict stability condition is

1>n(n−1)Δ~(F)+1n(n−1).1>n(n-1)\widetilde{\Delta}(F)+\frac{1}{n(n-1)}.

Substituting n=r+1n=r+1 into the criterion gives

n(n−1)Δ~(F)+1n(n−1)=rdg−1+1−r3r(r+1)<r+1−r3r(r+1)=r2+1r(r+1)<1,n(n-1)\widetilde{\Delta}(F)+\frac{1}{n(n-1)} =\frac{rd}{g-1}+\frac{1-r^3}{r(r+1)} <r+\frac{1-r^3}{r(r+1)} =\frac{r^2+1}{r(r+1)}<1,

where the strict inequality uses d<g−1d<g-1. Thus Theorem 4.4 implies that F∣CF|_C is μL∣C\mu_{L|_C}-stable. □

Universidad Autónoma de Aguascalientes

Av. Universidad # 940, Ciudad Universitaria,

C.P. 20100, Aguascalientes, Aguascalientes, México.

Email address: [email protected]

SECIHTI-ITCM

1° Mayo & Sor Juana Inés de la Cruz, Los Mangos.

C.P. 89440 Cd. Madero, Tam. México

Email address: [email protected]

References

  1. [1]Arend Bayer. Wall-crossing implies brill-noether. applications of stability conditions on surfaces. arXiv preprint arXiv:1604.08261, 2016.
  2. [2]Arend Bayer and Chunyi Li. Brill–noether theory for curves on generic abelian surfaces. Pure and Applied Mathematics Quarterly, 13(1):49–76, 2018.arxiv.org/abs/1704.03546
  3. [3]Soheyla Feyzbakhsh. An effective restriction theorem via wall-crossing and mercat’s conjecture. Mathematische Zeitschrift, 301(4):4175–4199, 2022.arxiv.org/abs/1608.07825
  4. [4]Daniel Huybrechts. Complex geometry: an introduction. Springer, 2005.DOI
  5. [5]Daniel Huybrechts and Manfred Lehn. The geometry of moduli spaces of sheaves, volume 2. Cambridge University Press, 2010.
  6. [6]Andreas Leopold Knutsen, Margherita Lelli-Chiesa, and Giovanni Mongardi. Severi varieties and brill–noether theory of curves on abelian surfaces. Journal für die reine und angewandte Mathematik (Crelles Journal), 2019(749):161–200, 2019.DOI
  7. [7]Robert Lazarsfeld. Brill-noether-petri without degenerations. Journal of Differential Geometry, 23(3):299–307, 1986.DOI
  8. [8]Margherita Lelli-Chiesa. Stability of rank-3 lazarsfeld–mukai bundles on k3 surfaces. Proceedings of the London mathematical society, 107(2):451–479, 2013.DOI
  9. [9]Federico Moretti. Brill–noether theory and green’s conjecture for general curves on simple abelian surfaces. International Mathematics Research Notices, 2025(7):rnaf073, 2025.arxiv.org/abs/2205.15977
  10. [10]Poornapushkala Narayanan. On the semistability of certain lazarsfeld–mukai bundles on abelian surfaces. ANNALI DELL’UNIVERSITA ’DI FERRARA, 64(1):145–164, 2018.DOI

Paper details

vlnc:W_xyawysdu
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents